This could be the same as recording a sample, but instead of using the audio device input it should use input from some pattern track. Such feature will be great for recording live VST improvisations that often get screwed by the limited time resolution if recorded as pattern data, especially when playing fast parts.
Another approach to this would be having “Render selection to sample” recording in real time with the audio still audible so that one can play along and tweak things live.
Either of these will do nicely for me, though I’m a bit more inclined to the first. Anyone else?